Access Selection and Pricing in Multi-Operator Wireless Networks

(1)

HAL Id: tel-01521312

https://tel.archives-ouvertes.fr/tel-01521312

Submitted on 11 May 2017

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

Access Selection and Pricing in Multi-Operator Wireless

Networks

Soha Farhat

To cite this version:

Soha Farhat. Access Selection and Pricing in Multi-Operator Wireless Networks. Networking and Internet Architecture [cs.NI]. Universite de Rennes 1, 2016. English. �tel-01521312�

(2)

THÈSE / UNIVERSITÉ DE RENNES 1

sous le sceau de l'Université Bretagne Loire En cotutelle internationale avec

l'Université Libanaise, Liban

pour le grade de

DOCTEUR DE L'UNIVERSITÉ DE RENNES 1

Mention : Informatique

Ecole doctorale MATISSE

présentée par

Soha FARHAT

préparée à l'IRISA (UMR 6074)

Institut de Recherche en Informatique et Systèmes Aléatoires

Access Selection

and Pricing

in Multi-Operator

Wireless Networks

Thèse soutenue à Rennes le 19 Juillet 2016

devant le jury composé de : Tijani CHAHED

Professeur à l'Université Telecom SudParis, FRANCE / rapporteur

Zaher DAWY

Professeur à l'Université Américaine de Beyrouth, LIBAN / rapporteur

Véronique VEQUE

Professeur à l'Université de Paris Sud, FRANCE / examinatrice

Bernard COUSIN

Professeur à l'Université de Rennes 1, FRANCE / directeur

Abed Ellatif SAMHAT

Professeur à l'Université Libanaise, LIBAN / directeur

Samer LAHOUD

Maitre de Conférences à l'Université de Rennes 1, FRANCE / co-directeur

(3)

(4)

Acknowledgements

Firstly, I would like to express my sincere gratitude to my two supervisors, Prof. Abed Ellatif SAMHAT, and Dr. Samer LAHOUD for the continuous support of my Ph.D study and related research, for their patience, motivation, and immense knowledge. Their guidance helped me in all the time of research and writing of this thesis. I could not have imagined having a better advisors and mentors for my Ph.D study.

My sincere thanks also goes to Pr. Bernard COUSIN, who provided me an opportunity to join his team, and who gave me support during my stay at IRISA Laboratory.

Besides, I would like to thank Prof. Tijani CHAHED, Prof. Zaher DAWY, and Pr. Veronique VEQUE, for accepting to be a part of my thesis committee.

I would like to thank all my colleagues at IRISA, especially Melhem, Mohamad, Farah and Nadia, and all of my friends who supported me in writing, and incented me to strive towards my goal, many thanks to Imen, Rida, Rima, Soukayna and Hassan.

I would like to thank my colleagues at the Faculty of Engineering of the Lebanese Uni-versity, Mr. Amine HAYDAR, Mr. Khayrallah ALFAKIH and Rami for their technical support.

Last, but not the least, special thanks to my family. Words cannot express how grateful I am to my mother, for all of the sacrices that you've made on my behalf. Your prayer for me was what sustained me thus far. I am extremely grateful to my sisters and my husband for all caring support during the nalization of this dissertation.

(5)

Résumé

La cinquième génération de réseaux mobiles, 5G, est destinée à prendre en charge le besoin croissant en bande passante, l'accroissement du nombre de mobiles connectés à des équipements et l'évolution des services attendus par les usagers. Il est prévu que la 5G fournisse une capacité beaucoup plus grande que la quatrième génération (4G) pour répondre à la demande croissante des utilisateurs, suite à l'apparition de multiples nouveaux services. En fait, le volume de données échangé devrait d'ailleurs être mul-tiplié par 1000 avec le nombre croissant de terminaux connectés. La 5G a aussi pour objectif de permettre l'explosion attendue de l'internet des objets, accompagnant les nouveautés comme les villes intelligentes, les voitures sans conducteur ou les systèmes de soins de santé. On envisage un grand nombre de capteurs, machines industrielles et de transport connecté ayant besoin d'une connexion ubiquitaire et à tout moment. La 5G devra être un réseau mobile à ultra haut débit et peu consommateur en ressources énergétiques. Diérentes technologies se complèteront mutuellement pour atteindre les objectifs de la 5G. En fait, la densication des antennes du réseau est un des moyens pour renforcer la capacité des réseaux mobiles contre la croissance du trac. De plus, le déploiement de petites cellules comme des métrocellules, des picocellules et des fem-tocellules, présente une solution économique permettant d'accroître encore la capacité et réduire la consommation d'énergie, grâce à des modalités intelligentes d'orientation et de délestage du trac. En outre, l'exploitation de bandes de fréquences plus élevées, les techniques de non-orthogonalité et les antennes multiples, associés au partage du spectre, sont des facteurs clés pour parvenir à une plus grande ecacité spectrale. Le passage à la 5G imposera des changements non seulement dans le réseau d'accès radioélectrique mais aussi dans le réseau central, où les solutions logicielles joueront un rôle essentiel pour assurer la connectivité à un nombre croissant d'utilisateurs et de dispositifs. La tendance actuelle est de découpler le matériel du logiciel et de faire migrer les fonctions du réseau vers le logiciel, an de réaliser une séparation entre la commande et les données. Ainsi, des eorts de normalisation visent à dénir la virtualisation des fonctions du réseau. En conséquence, avec une exploitation plus simple, de nouvelles caractéristiques du réseau seraient déployées plus rapidement. Dans certains pays, la 5G devrait en eet être lancée commercialement pour 2020. C'est pourquoi, les opérateurs mobiles devraient continuer d'investir, dans les prochaines années, dans le déploiement de leurs réseaux mobiles à très haut débit, qui vont leur permettre d'augmenter les débits et d'adapter la capacité des

(6)

réseaux à la hausse exponentielle du trac. En eet, pour déployer ce futur réseau mobile, beaucoup d'argent a déjà été mis sur la table principalement chez Huawei et Samsung, par la Commission Européenne, par la partie privée du 5GPPP et par la Corée du Sud. Dans ce contexte, certains régulateurs comme l'Arcep trouve que les accords de partage de réseaux mobiles peuvent constituer pour les opérateurs un moyen d'accélérer et de ré-duire les coûts de déploiement tout en améliorant leur ore de services. Ainsi, parmi les caractéristiques essentielles des futurs réseaux mobiles, on compte une capacité accrue, de moindres dépenses d'investissement et d'exploitation, une ubiquité complète assurée par un interfonctionnement multinorme ainsi que le partage du spectre et de l'infrastructure. Le partage de réseaux mobiles consiste à mettre en commun entre plusieurs opérateurs tout ou partie des équipements constituant leurs réseaux mobiles. On distingue deux grands types de partage d'infrastructures actives : l'itinérance qui consiste en l'accueil, par un opérateur de réseau mobile, des clients d'un autre opérateur de réseau mobile sur son réseau, pour lequel seules les fréquences de l'opérateur accueillant sont exploitées. Et, la mutualisation des réseaux qui contrairement à l'itinérance, exploite les fréquences des deux opérateurs. Notre travail se situe dans le contexte de partage de réseau mobile actif, ou un nombre d'opérateurs partagent leur accès radio, an de former un system multi-technologie multi-opérateur. Dans cet environnement coopératif, un utilisateur mobile peut être servi à travers le réseau de son opérateur de domicile, avec lequel il a fait un con-trat, ou il est transféré par son opérateur de domicile pour être servi à travers le réseau d'un autre opérateur coopérant. Ce dernier déterminera le coût de transfert, qui sera payé par l'opérateur du domicile de l'utilisateur. Le but de notre étude est de montrer les avantages de la coopération entre les opérateurs, principalement en ce qui concerne les revenus. De plus, nous cherchons des stratégies pour surpasser les conséquences néga-tives du partage des ressources, surtout celles touchant la performance des réseaux des opérateurs coopérants. Nous avons montré que les bénéces de la coopération dépendent fortement du choix de partenaires, la tarication de service ( cout de transfert) entre les partenaires, et combien un opérateur partage de ses ressources.

Notre travail consiste, en premier temps, à proposer un algorithme de sélection d'accès applicable dans un réseau multi-opérateurs. Cet algorithme devrait garantir la satisfac-tion en QoS de l'utilisateur et celle en prot de son opérateur d'accès à l'Internet. Ainsi, un algorithme adoptant une décision hybride, NP-BPA (Nearest Performance and Best Prot Algorithm), est proposé. Il exploite la simplicité des algorithmes MADM (Mul-tiple Attribute Decision Making), spéciquement SAW (Simple Additive Weighting), et l'ecacité de l'approche NPH (Nearest Performance Handover). Cet algorithme de sélec-tion est basé sur une foncsélec-tion de coût combinant les exigences du service de l'utilisateur mobile, et le prot résultant du transfert de l'utilisateur à un autre opérateur. La com-paraison de performance de notre algorithme, NP-BPA, avec d'autres méthodes MADM, comme SAW et NPH, a montré son ecacité concernant la probabilité de blocage et le prot global réalisé. Notre algorithme de décision garantit la plus faible probabilité de

(7)

blocage pour tous les opérateurs comme il évite les surcharges du réseau. En outre, NP-BPA donne aux opérateurs la possibilité d'exprimer leur stratégie lors de l'exécution de la sélection, et ainsi faire du contrôle d'accès tout en utilisant deux coecients explicites dans la fonction de coût.

En deuxième temps, nous étudions la tarication de service entre les opérateurs parte-naires, précisément le coût de transfert d'un utilisateur. Ce dernier paye juste le prix du service que son opérateur d'accès à l'Internet détermine, il est inconscient du transfert. Les modèles de tarication proposés relient le coût de transfert d'un opérateur au prix adopté pour le service des clients. Le premier modèle, ACAG (As Client As Guest), sug-gère que le coût de transfert d'un opérateur soit égal à son prix de service. Le deuxième modèle, MIWC (Maximum Income When Cooperating), suggère que les coûts de trans-fert des opérateurs coopérants soient identiques, et égaux au prix de service le plus élevé des partenaires. Et, le troisième modèle, MCWC (Minimum Cost When Cooperating), suggère que les coûts de transfert des opérateurs coopérants soient identiques et égaux au plus petit prix de service des partenaires. L'étude de la protabilité de ces modèles dans un system à trois opérateurs, et la comparaison au modèle de partage de prix, ont montré que nos modèles garantissent les prots les plus élevés, et assurent le partage de prot entre les partenaires en respectant leur capacité partagée. La décision du meilleur modèle à adopter lors de la coopération, intervient une interaction entre les diérents partenaires. Nous avons modélisé cette interaction à l'aide de la théorie de jeux. Nous avons exploité un jeu Stackelberg à deux niveaux, TPA (Transaction Pricing and Access Selection), où les opérateurs de service agissent comme Leaders et les opérateurs d'accès à l'Internet des utilisateurs à transférer agissent comme Followers.

Finalement, nous avons considéré le mode d'accès hybride pour la coopération. Ce mode d'accès est proposé comme solution surtout pour les opérateurs partageant la plus grande capacité. La performance du réseau de ces opérateurs est relativement aaiblie suite à la coopération. Nous avons vérié que le pourcentage de blocage diminue quand l'opérateur, ayant une capacité élevée, réduit le pourcentage de ressources partagées. Pour un même pourcentage de partage, le prot d'un opérateur dière avec le modèle de tarication adopté. Ainsi, une bonne décision doit être prise, concernant le pourcentage de partage et le modèle de tarication, tout en tenant compte de l'eet de cette décision sur les autres partenaires du système. C'est pourquoi que nous avons proposé un nouvel jeu séquentiel à deux niveaux, RS-TP (Resource Sharing and Transaction Pricing) an de modéliser l'interaction entre les opérateurs, pour le partage de ressources et la tarication du coût de transfert.

(8)

Abstract

5G networks will rely on virtualization and network sharing in order to address the explosive growth of broadband trac mobile, the increasing number of mobile connected devices and the evolution of mobile user expectations. Network sharing is a powerful approach that helps to reduce the deployment time and cost of a new radio access network, expand coverage, accelerate the integration of a new technology and to optimize resource utilization. Further, it is ecient for new revenues achievements.

We consider a roaming-based infrastructure sharing system, where multiple operators share their radio access in a multi-operator environment. In such system, mobile users can access the base station (BS) of their home operator or the base station of another partner of the sharing system. We assume that all BSs of the partners remain active and the users are not free to access another operator BS without the permission of their home operator. Indeed, when the home operator of a user is unable to satisfy its constraints, because of lack of resources or QoS, a transaction event is triggered. It consists in transferring the considered user to another operator in order to access the service. Moreover, when there are more than two operators sharing their access, the user transfer process includes an access selection decision in order to choose the best operator for service. We assume that the access selection decision is triggered and controlled by the home operator of the transferred user. Furthermore, when a user is transferred, its home operator must pay some transaction cost as cooperation fees for the new service operator. This transaction is seamless to the user. Therefore, the inter-operators sharing agreement set for cooperation must include three important issues: the selection decision algorithm, the transaction cost pricing scenario, and the percentage of resources shared by each operator.

In the rst part, we introduce our selection decision algorithm in a multi-operator environment, NP-BPA (Nearest Performance and Best Prot Algorithm). It is based on a multi-criteria cost function which groups the dierent parameters that enable a satis-fying selection decision, for the operators and users. In this decision process, the home operator of the transferred user is the main player, it triggers and performs the selection applying its own strategy using our cost function. We show the eciency of our selection algorithm in dierent environments considering dierent numbers of partners. Besides, the performance of NP-BPA algorithm was compared to MADM (Multiple Attribute De-cision Making) methods, precisely SAW (Simple Additive Weighting), and NPH (Nearest Performance Handover), in a three operator environment. NP-BPA showed better

(9)

re-sults for the blocking rates and global achieved prots. Our algorithm helps to reduce overloading situations for the service operators; its distributes the transferred users in an ecient manner and thus improves the prots for all cooperating partners.

In the second part, we study the transaction cost. We nd rational that an operator sets its transaction cost as a function of its service price. First, the service price is a public parameter that can be easily exchanged with other partners. Second, if an operator decides to vary its service price or to adopt dynamic pricing, it will aect directly the transaction cost. We consider a sharing system of three partners, interacting to decide the best transaction cost. Taking into account that the service of a guest user may aect the probability of acceptance of a client, an operator looks for preserving the expected revenue from its client. Therefore, we propose the rst pricing scenario, ACAG (As Client As Guest) that aims to set the transaction cost of an operator equal to its service price. However, every operator seeks to maximize its revenue; therefore it is expected to set a higher transaction cost. How much higher? This must respect the sharing agreement between dierent partners and the service prices they adopt. To be optimistic, we propose a second pricing scenario MIWC (Max In When Cooperating). With this scenario all partners agree to have a transaction cost equal to the highest service price announced in the system. But, this scenario may cause losses in some cases where an operator setting a low service price performs a lot of transactions. To be fair, we propose a third pricing scenario MCWC (Min Cost When Cooperating). With this scenario all partners agree to have a transaction cost equal to the lowest service price announced in the system. Although this pricing scenario seems hypothetical, it is more protable than ACAG, in some systems.

Next, to study the protability of these pricing scenarios we presented two system models: In the rst system, the operators set the same pricing scenario but share dif-ferent capacities. In the second system, the operators share the same capacity but set dierent service prices. In both systems we compare the achieved prots using our pricing scenario and price sharing scenarios. Results show that the best pricing scenario for an operator depends on its shared capacity and the service price it sets. Besides, one pricing scenario may maximize the prots of one operator but not of the others. Hence, to decide the best pricing scenario to adopt in the sharing system, a two stage Stackelberg game, TPA (Transaction Pricing and Access Selection) game, is formulated. In this game, the operators are the players; the service operators are the leaders and the home operator of a transferred user is a follower. Two cases were studied: the rst one where all operators adopt the same pricing scenario. In this case we found the U-TPA (Uniform Transaction Pricing and Access Selection) equilibrium. And, the second case where each operator adopts its own pricing scenario. In this case we found the F-TPA (Free Transaction Pric-ing and Access Selection) equilibrium. In both cases the equilibrium scenario is MIWC. In fact, in the system where the partners share dierent capacities and set dierent service price, MIWC guarantees the best prot sharing among all partners.

(10)

In the third part, we consider a three operator sharing system with hybrid access mode. In this system partners decide to share a restricted amount of their capacity. We show how the sharing factor aects the blocking rates and aect the global prots. Further, the achieved prot does not depend only on the sharing factor, but also on the adopted pricing scenario. Therefore an economic framework based on game theoretical analysis is proposed. It models the interaction between the sharing system operators for resource sharing and pricing, in addition to the access selection. A sequential game is formulated, RS-TP game (Resource Sharing and Transaction Pricing), where the players are the operators. In the rst stage, the sharing partners decide the proportion of resources they will share and the transaction pricing scenario in order to maximize their own prots. In the second stage, the home operator of a transferred user selects the suitable service operator. A bi-level optimization problem is solved and equilibrium is found.

(11)

List of Figures

2.1 RAN Sharing Levels[PO14] . . . 7

2.2 MADM with AHP-based weighting . . . 13

2.3 An overview of the main modules of the Flex access market . . . 15

3.1 System Logic . . . 22

3.2 Decision parameters . . . 24

3.3 System Model . . . 27

3.4 Global blocking rates-CEC case . . . 28

3.5 Op3's and Op3x's blocking rates-CEC case . . . 29

3.6 Op3's global prots-CEC case . . . 30

3.7 Op3x's global prots-CEC case . . . 30

3.8 Global blocking rates-CLC case . . . 31

3.9 Op1's network blocking rates-CLC case . . . 31

3.10 Op3's network blocking rates-CLC case . . . 32

3.11 Op1's global prots-CLC case . . . 33

3.12 Op3's global prots-CLC case . . . 33

3.13 Global blocking rates-CHC case . . . 34

3.14 Op2's network blocking rates-CHC case . . . 35

3.15 Op3's network blocking rates-CHC case . . . 35

3.16 Op2's global prots-CHC case . . . 36

3.17 Op3's global prots-CHC case . . . 36

3.18 Global blocking rates in Sys3 . . . 38

3.19 Operators' network blocking rates in Sys3 . . . 38

3.20 Operators' Global Achieved Prots in Sys3 . . . 40

3.21 Global blocking rates in Sys4 . . . 41

3.22 Operators's Networks blocking rates in Sys4 . . . 42

3.23 Operators' Global Prots . . . 43

3.24 Comparison of the global blocking rates . . . 44

3.25 Comparison of operators' networks blocking rates . . . 46

3.26 Comparison of operator's global prots . . . 47

3.27 Comparison of Op1's income decomposition and costs . . . 48

(15)

List of Figures

3.29 Comparison of Op3's income decomposition and costs . . . 50

4.1 Operators's Achieved Prots in BSBC . . . 59

4.2 Partners' prots ratio in BSBC . . . 60

4.3 Operators's Achieved Prots in BSBP . . . 61

4.4 Transaction Pricing and Access selection Game Hierarchy . . . 63

4.5 Equilibrium scenario for each pair of players . . . 65

4.6 Partners' Prot ratio-General case . . . 66

5.1 Op1's Blocking Rates Comparison with static sharing . . . 74

5.4 Op1's Achieved Prots with static sharing . . . 78

5.5 Eect of the sharing factor and pricing scenario on the achieved prots of the partner sharing the highest capacity, Op2 . . . 79

5.6 Eect of the sharing factor and pricing scenario on the achieved prots of Op3 80 5.7 RS-TP game hierarchy . . . 82

5.8 Extensive form representation of the low level (followers) game and subgames 1&2 . . . 87

(16)

List of Tables

2.1 Network Sharing Deals in Europe . . . 11

2.2 Decision parameters and their utility functions . . . 12

2.3 Comparison of Network Selection Techniques . . . 16

3.1 Operators' Delivered Parameters . . . 27

3.2 Serving rates of guest users-CEC . . . 29

3.3 Serving rates of guest users-CLC . . . 32

3.4 User's Application Requirements . . . 37

3.5 Op2's Serving rates (%) in Sys3 . . . 39

3.6 Comparison of the selection percentages of the service operators(%) . . . 44

3.7 Candidate operators qualication . . . 52

3.8 Service Operator selection(%) . . . 52

4.1 Operators' networks parameters and service prices of Sys3 . . . 57

4.2 The delivered parameters and service price of Sys3 . . . 64

4.3 Op1, Op2 and Op3 Payos when Op1 chooses ACAG . . . 68

4.4 Op1, Op2 and Op3 Payos when op1 chooses MIWC . . . 69

4.5 Op1, Op2 and Op3 Payos when op1 chooses MCWC . . . 69

4.6 Op1, Op2 and Op3 Payos when op1 chooses pShareL . . . 70

5.1 Partners' delivered parameters and service prices . . . 72

5.2 Game output payos . . . 87

(17)

Nomenclature

3GPP 3rd Generation Partnership Project

5G Fifth Generation

APs Access Points

BBU Baseband Processing Units

BS Base Station

C-RAN Cloud Radio Access Networks

CAPEX CAPital EXpenditure

CC-CRRM Cloud-Computing based Cooperative Radio Resource Management

CRRM Coordinated Radio Resource Management

D2D Device-to-Device

ESS Evolutionary Stable Strategy

F-TPA Free Transaction Pricing and Access Selection

H-CRAN Heterogeneous Cloud Radio Access Networks

H-op Home operator

HetNet Heterogeneous Networks

MADM Multiple Attributes Decision Making

MIMO Multiple-Input Multiple-Output

MNOs Mobile Network Operators

MSP Macro Cell Service Provider

(18)

List of Tables

NFV Network Function Virtualization

NP-BPA Nearest Performance and Best Prot Access

OPEX OPerational EXpenditure

RAN Radio Access Networks

RS-TP Resource Sharing and Transaction Pricing

S-op Service operator

SCNs Small Cell Networks

SCSP Small Cell Service Provider

SDN Softaware Dened Network

SLA Service Level Agreements

SPs Service Providers

TPA Transaction Pricing and Access Selection

(19)

Chapter 1 Introduction

1.1 Motivations

Fifth generation mobile networks must address new challenges that appeared with the explosive growth of the mobile trac broadband, the increasing number of mobile con-nected devices and the evolution of mobile user expectations. In fact, global mobile data trac grew 74 percent in 2015 and it is expected to increase nearly eightfold between 2015 and 2020 [Ind16]. In addition, mobile users are more aware of the QoS and are evaluat-ing increasevaluat-ingly the connectivity, especially for the services with high QoE expectations. The need of high-speed connectivity for anything, anywhere and anytime is growing, and operators are facing the challenge to upgrade their network in order to expand capacity, support higher data rates and enhance QoS in terms of End-to-End (E2E) latency with energy and cost eciency. Further, the growth of data consumption is overtaking voice usage [Eri15, Mar11], thus aecting operator's revenues. Consequently, new strategies are needed for new network deployment or the rollout of 5G technology, that helps operators to keep up with the mobile market and ensure additional incomes.

5G mobile technology promises innovation for entire mobile industry [HUA13, 5GP15, NGM15]. It targets massive capacity and connectivity in order to support an increasingly diverse set of services, applications and users with extremely diverging requirements. It aims for a exible and ecient use of available radio resources. Future mobile networks will adopt new solution frameworks to accommodate both LTE and air interface evolution, as Cloud, SDN and NFV technologies. The roadmap of 5G includes a number of network

and technology solutions as [YOU15, AIS+_{14, 5GP15]:}

• The use of millimiter-wave frequencies and Massive MIMO for maximum

transmis-sion data rates 20 times as fast as 4G LTE.

• The use of full duplex radio technologies and Device-to Device (D2D), in order to

improve the downlink spectral eciency.

(20)

1.2. Problem Statement

• The use of ecient inter-cell interference management for ultra-dense networks.

• Radio Access network Virtualization in a Cloud-based radio access infrastructure

[GiC16][CCY+15].

An important step in dening 5G has already been made in the Next Generation Mobile Networks (NGMN) [NGM15] where 25 use cases have been identied, grouped into eight use case families, and serve as input for specication of requirements and dening the building blocks of the 5G architecture.

For operators, time and cost are crucial. Therefore, a rational decision have to be done in order to hold on with the mobile market evolution. And, since the growth of trac and revenues are decoupled, new sources of revenues and new cost reduction solutions are needed. RAN (Radio Access Network) sharing is a rational approach that can help to reduce costs, to maximize eciency and competitiveness, and to enhance customer satisfaction. It is introduced as a cost eective solution to expand coverage and increase capacity in [Cor13, Joh07, FSL14, JG13]. It involves active sharing of RAN between two or more operators as a mean of mutually oering access to each other's resources. This inter-operators arrangement brings a lot of benets for operators as CAPEX and OPEX savings, new revenues achievements and energy consumption reduction. Besides, it promotes innovation since the competition between operators, in such environment, is based on oered services and features [MGM13]. In fact, current 3rd Generation Partnership Project (3GPP) standards fully support network sharing between operators under dierent sharing scenarios as Multi-Operator Core Network (MOCN) and Gateway Core Network (GWCN) [rGPP13].

Nowadays, a key factor for achieving infrastructure sharing is the virtualization of physical entities by decoupling their functionality from the hardware. Further, network densication and small cell deployment are achievable through virtualization in H-CRAN; femtocells and picocells are created by RRHs instead of low power base stations (BS) and access points, the infrastructure workload is computed at the BBU, where resource

avail-ability as well as overloading of physical entities becomes easier to assess [MKGM+_15].

1.2 Problem Statement

We consider an infrastructure sharing system, where multiple operators share their radio access in a multi-operator environment. This environment includes RAN virtualization and also a Cloud based radio access infrastructure. In such system, mobile users can access the BS of their home operator or the BS of another partner of the sharing system. We assume that all BSs of the partners remain active and the users' access to another operator BS is controlled by the home operator. Indeed, when the home operator of a user is unable to ensure its satisfaction constraints, because of lack of resources or QoS, a transaction is triggered to transfer the considered user to another service operator and

(21)

1.3. Thesis Organization the access to the service is granted through the network of this operator, thus avoiding the user rejection.

Accordingly, every operator have to adopt a suitable strategy for serving the users of another operator (guest users) without aecting its network performance or its own subscribers satisfaction. Actually, the hybrid access mode is the most promising because it allows operators to give preferential access to their own subscribers, while other guest users can only access a restricted amount of resources. Besides, when there is more than two operators sharing their access, the transaction process includes an access selection decision in order to choose the best operator to serve the user. The access selection decision is triggered and controled by the home operator of the transferred user. In fact, in a multi-operator environment, a hybrid approach for the selection decision is a need, in order to guarantee the user satisfaction and the operators happiness in the same time, especially when considering the cost to pay for the transaction. In fact, when transferring the user to a new service operator, some transaction fees must be paid to this operator. The home operator has to make this payment, in order to keep this transaction seamless to the user.

Therefore, the inter-operatorss sharing agreement set for cooperation must include three important issues:

1. The selection decision algorithm adopted in the sharing system.

2. The transaction fees, characterized by a transcation cost for each operator. 3. The percenatge of resources shared by each operator.

We assume that a third trusty party is integrated in order to maintain and guarantee the inter-operators agreements especially for the transaction cost pricing.

1.3 Thesis Organization

The remaining of this thesis is organized as follows: A survey on RAN sharing is in-troduced in Chapter 2. We discuss the benets of a RAN sharing and reveal the main challenges of a successful sharing agreement. In addition, we investigate the radio access selection decision in a multi-operator environment, and classify a wide range of meth-ods, using simple mathematical tools, including Multiple Attributes Decision Making (MADM), Fuzzy and Game Theories.

Moreover, dierent strategies for the inter-operators service pricing are represented, it includes the dierent business models that may be adopted during cooperation and how to determine the inter-operators service cost between sharing partners. Further, an overview of dierent modeling frameworks for the access selection and service pricing is made. These models use game theory in order to describe the interaction between the service providers in a multi-operators network.

(22)

1.3. Thesis Organization Chapter 3 introduces our selection decision algorithm, NP-BPA, in multi-operators environment. It is based on a cost function which considers jointly the oered QoS pa-rameter oered by the service operators and the prot of the home operator resulting from the transaction. The performance of this selection algorithm is investigated in dier-ent sharing environmdier-ents considering dierdier-ent numbers of partners. Then a performance comparison is made with MADM methods, precisely SAW and NPH, in a three opera-tors environment. Further, an analysis of two coecients of the cost function reveals the ability of an operator to express its strategy and to control the access selection decision of its user.

In Chapter 4, the inter-operators service cost is studied and three basic pricing scenar-ios are proposed. These pricing scenarscenar-ios determine the transaction cost of an operator as a function of its service price or the service price of other partners in the sharing system. Further, the protability of these scenarios are compared with other pricing scenarios in litterature that consist of sharing the transferred user payment between its home oper-ator and the new service operoper-ator. Moreover, the decision of the best pricing scenario to be adopted in the system is achieved using game theory; the interaction between the operators of the sharing system is modeled using a Stackelberg game where the available service operators are the leaders and the home operator of a transferred user is a follower. Chapter 5 grabs resource sharing and reservation in a three operator system. It shows how resource reservation can guarantee client satisfaction by reducing the blocking rates. In addition, the inter-operators service pricing and the protability of the previously pro-posed scenarios are investigated in a hybrid access mode. Further, an economic framework based on game theoratical analysis is proposed. The framework formulates a two-stages sequentiel game in which the sharing partners decide the proportion of resources they will be shared with other partners, and the transaction pricing scenario to adopt in order to maximize their prots.

Chapter 6 concludes the thesis, where we summarize the main contributions, and present future research directions.

(23)

Chapter 2 Access Selection and Service Pricing in

Multi-Operator Shared Networks

With the exploisive growth of mobile broadband trac, the MNOs must consider new measures to upgrade their networks in order to expand coverage, increase capacity and enhance service quality with cost reduction optimization. Network sharing is a powerful approach to bring down network costs on both relative and absolute scale. It involves RAN and networking infrastructure sharing between two ore more mobile operators. In such sharing networks, an agreement must be set between operators for the access selec-tion decision process, the inter-operators service cost and the resource sharing policy to be adopted during cooperation. This chapter briey introduces the motivations of RAN sharing, the benets and the challenges to achieve a successful network sharing transca-tion. And, it presents some exemples of current RAN sharing markets. Furthermore, an overview is made on the access selection decision making approaches, in single and multi-operator heterogeneous networks. Moreover, dierent strategies for the inter-multi-operators service pricing are represented, it determines the transaction cost between partners.

2.1 Why RAN Sharing?

5G mobile networks must address new challenges that appeared with the explosive growth of the mobile trac broadband, the increasing number of mobile connected devices and the evolution of mobile user expectations. In fact, global mobile data trac grew 69% in 2014 [Ind15], and it is expected to increase nearly eightfold between 2015 and 2020 [Ind16]. Besides, the need of high-speed connectivity for anything, anywhere and anytime is growing, and operators are facing the challenge to upgrade their network in order to expand coverage, increase capacity, support higher data rates and enhance QoS in terms of E2E latency, with energy and cost eciency.

In addition, for Mobile Network Operators (MNOs) the speed of new technologies introduction, the quality of the network and indoor coverage are main factors that

(24)

in-2.1. Why RAN Sharing? uence a mobile customer decision for the choice of an operator and his willigness to pay for access. Hence, it is necessary to nd a cost ecient solution in order to achieve an optimal balance between prots and costs. Network sharing is a powerful approach that can help to accelerate coverage expansion, reduce deployment period and optimize resource utilization. Further, it is ecient for additional CAPEX and OPEX savings and new revenues achievements.

Network sharing consists of RAN sharing, i.e, the radio access layer which contains the infrasctructure and the base station subsystem, between two or more MNOs. Typically, RAN represents the one-third of the total OPEX and 80% of CAPEX, and it counts 52% of total indirect network costs [HDT09]. RAN sharing arrangement brings a lot of benets for operators and it promotes innovation since the competition between operators, in such environment, is based on oered services and features. RAN sharing is a promotive approach when a MNO has already reached the limits of cost improvement. In addition, it is advantageous for operators who seek new investments, as well as in greenelds situations where new technology could require a rethinking/renewal of the network insfrastructure [GSM15].

5G networks will likely rely on RAN sharing [Net15, ASD15], in order to accelerate and reduce the cost of new RAN deployment using, for instance, new millimeter wave spectrum, in addition to sophisticated multi-tower, multi-carrier aggregation.

2.1.1 What to share?

Currently, network sharing is mainly conned to elements of the RAN such as infrastruc-ture and base station subsystem elements [GSM15]. Few sharing deals do include parts of the core networks and spectrum because of regulatory issues that aim to maintain net-works capabilities dierentiation. Besides the cost benets of core netnet-works sharing are not as great as the benets of RAN sharing [GSM15]. Current 3GPP standards fully sup-port network sharing between operators [rGPP13, Net15], it denes three dierent levels determining how shared networks are integrated. The diagrams in Fig. 2.1 describes the dierent sharing levels:

1. Multi-Operator RAN (MORAN) sharing is where only equipments are shared. 2. Multi-Operator Core Network (MOCN) sharing is where both equipments

and spectrum are shared.

3. Gateway Core Network (GWCN) sharing is equipments, spectrum and some core network elements are shared.

In practical, operators do not share the entire RAN, and can maintain dedicated RAN in the areas where trac is heavy. Some mobile economics analyst nd that RAN sharing is ecient in the markets where most customers have pre-paid service, thus, having more network availability means more billable minutes, thus more revenues.

(25)

2.1. Why RAN Sharing?

Figure 2.1: RAN Sharing Levels[PO14]

The partnership structure identies the dimensions of the shared network. According to RADAR approach [HDT09], the sharing network has four dimensions:

1. Geography: It determines which locations will be shared urban, rural, selected urban and rural or countrywide.

2. Infrastructure: It determines the physical components of the network to be shared, with two sharing categories:

• Passive RAN sharing: where operators share only physical cell sites and

tow-ers and passive infrastructure elements like shelttow-ers, masts, air conditionning and power supplies.

• Active RAN sharing: where operators share passive equipments as well as

transport infrastructure (radio access nodes and transmission), radio spectrum and baseband processing resources. Generally, the RAN sharing is not uniform, passive sharing may be used in some locations and active sharing in others.

(26)

2.1. Why RAN Sharing? 3. Technology: It determines which mobile capabilities are to be shared 2G, 3G or 4G technology. MNOs might share some combination of these technologies or all of them.

4. Process: its determines the services to be shared as:

• Engineering, planning and design.

• Deployment and rollout.

• Optimization.

• Maintainance and operation.

2.1.2 How to share?

How to share determines the adopted structure for network sharing, it helps to specify the commercial, technical, operational and legal conditions of partnership. Three structures can be used [HDT09]:

1. New network: This structure is ideal when rolling out a new network generation, sharing partners build a new network together and share it.

2. All-in-one network: This is a non classical network sharing structure, where a MNO provides the network while the others abondon their networks and benet from wholesale network services from recipient MNO, which may include national roaming and Mobile Virtual Network Operator (MVNO) services. In this structure, the parties agree that one operator will build, own and operate the network in one geographical area and allow others to roam, with the same arrangement in reverse in another geographic area.

3. Consolidated network: This structure arise where operators merge their net-works and deconstruct the redundant sites. In such structure the asset ownership may be handled by three forms:

• Joint Venture: The ownership is shared in a joint venture, which takes

the form of a common company that owns, operates and maintains the joint network. The parent MNOs contribute nancial and human resources to this joint venture. The most common structure adopted is the 50/50 joint venture [Hen14]. In fact, when a joint venture is formed for sharing, the operators are almost like MVNOs on the shared network.

• Third Party outsourcing: Where sharing operators transfer their assets

and outsource the management and operation of their shared network to a third party. In fact, 25% of the operators entrered this kind of arrangement,

(27)

2.1. Why RAN Sharing? although it reduces the savings and results in (Service Level Agreements) SLA-driven control of the third party as well as loss of competence with the oper-ator's organisations.

• Network Company: It is a service company where one operator is the owner

of the total network while the others pay for the service.

2.1.3 Benets of Network Sharing

The rst sharing agreement was made in Sweeden between Telia and Tele2, in early 2001. Telia was unable to acquire a 3G licence, so a joint venture on a 50-50 basis was established with here competitor Tele2, and it was able to become a 3G operator without having its own license. RAN sharing occurs as the best option for medium and small sized operators, as well as new operators [Hua11]. It reduces the network deployment period and accelerate the rollout of new technology in order to meet the time frames imposed by regulators. In addition, RAN sharing brings a lot of nancial advantages for MNOs [Hen14], it is able to:

1. Reduce the total cost of network ownership dened as the sum of costs to buy, to install, to operate and to maintain a network. In fact, sharing the access layer brings a lot of savings in CAPEX and OPEX. Through sharing, operators are able to save money to nd the appropriate site, to deploy the new site, to buy transmission and radio equipments, in addition to maintenance and power costs reduction.

2. Increase revenues resulting from widen service coverage, and wholesale arrange-ments which boost the return on capital.

3. Promote better utilisation of the network resources, spectrum pooling grants higher spectrum bandwidth and higher data thoughput.

4. Reduce the number of communication towers which scales down the environmental impacts [GSM15] for a green communication.

2.1.4 Challenges of Network Sharing

Four main challenges are to be considered in order to achieve a successful network sharing transaction [Hen14]:

1. Loss of independency: When an operator is engaged in a sharing arrangement he risks to lose independency of :

• The network operation (handover, performance KPIs and baseband capacity

split ratio...).

(28)

2.2. RAN Sharing Market

• The rollout strategies and vendor choices.

• The competitive developments with sharing partners in terms of service

dif-ferentiation.

2. Partner selection: With who to share is a very important issue, an operator must consider the following:

• The number of partners which aects the amount of cost savings in a shared

network and the geographical distribution of partners' sites which may cause additional costs of dismantling redundancies in a large overlap.

• The potentiel of dierentiation with other partners. In fact, the sharing

ar-rangement with a partner with a similar network is easier to reach. On the other hand, sharing with smaller or less advanced operator may cause loss of signicant competitive advantages.

• The alignment on network evolution and deployment, and investment plans

and strategies with the new partners.

3. Regulatory issues: Usually, the degree to which network sharing is allowed and supported by regulators diers by country. Mainly, regulators are concerned to maintain competitive dierentiation capabilities and avoid collision. These concerns are generally muted over passive sharing. Network sharing is allowed sometimes for environmental reasons and regulators tend to be more in favor of sharing in rural coverage.

A good sharing legal agreement must detail which entity has a full control over the whole network, how to evaluate assets and how to transfer existing assets into a joint venture structure. In addition, partners must agree on pricing transfer for ongoing services, how revenues are distributed and how operational, rental and power costs are shared.

2.2 RAN Sharing Market

Network sharing is not new, it has been started with national roaming and bilateral site sharing. Since 2001, three trends have emerged. Firstly, network sharing joint ventures between mobile network operators in Sweeden-Europe [Bui15]. A second trend, towerco deals, started in the last six years [Bui15]. It is the most frequent form of sharing around the world, where an operator sells its towers to a third party-or forms a joint venture- and leases them back, the majority of these deals have been in Africa and are taking place in other regions in the world especially in the Americas. A more recent trend is operator consolidation [Bui15]. And, the fourth trend that may emerge the next ve years is core network sharing [Bui15], considering the technology developments arising with network function virtualisation and software-dened networking.

(29)

2.2. RAN Sharing Market The network sharing picture in the Americas has been dominated by 70% of the towerco deals, in Brazil, Chile, Colombia, Mexico and the USA. The only active sharing deals between operators to date have been in Canada and Brazil. Europe is in the rst place with 15 active sharing deals but with only three towerco deals to 2015, in France, Spain and Netherlands. Africa leads the world in towerco deals with three multinational operators accounting for more than 80% of the deals. Asia Pacic stands out for its passive sharing between MNOs, operators are engaged in multiple deals as in India and Bangladesh. Some of network sharing deals in Europe, presented in [Bui15] are given in Table 2.1.

(30)

2.3. Access Selection Decision Making in Heterogeneous Networks

2.3 Access Selection Decision Making in

Heterogeneous Networks

2.3.1 Access selection decision in a Single Operator network

Access Selection was widely studied in heterogeneous wireless networks managed by a sin-gle operator. Various mathematical approaches that can be employed for access selection are presented and evaluated in [WK13]. Access selection tools include: utility and cost function used in [NVGDA08, OMmM06, BL07, KJ12b], Multiple Attribute Decision Mak-ing (MADM) methods in [KJ12b, SJ05, Zha04, HLIK13, SNW06, MMPRSN10, KJ12a, HILK13], Fuzzy Logic in [Zha04, GAPRS05, GAPRS09], Markov Chain in [SJ06, HILK14] and Game Theory in [CSMW02, ZBDH14, SPTC15, AHNK11, AP07, CMT08].

In a cost function based algorithm, decision parameters are normalized, assigned a weight and then injected into a weighted sum to produce a selection score. The decision parameters used for access decision and their utility functions are resumed in [WK13] and represented in Table 2.2. We can distinguish the bandwidth, BER, the delay, the jitter, the price and latency, used with Linear and sigmoidal utility functions.

Table 2.2: Decision parameters and their utility functions

Attribute Utility Functions

Bandwidth Linear, logarithmic, sigmoidal

Battery Linear

Price Linear, logarithmic

latency Linear

Interruption Probability Linear

Trac Linear, sigmoidal

Power Consumption Linear

BER Linear, sigmoidal

Delay Linear, sigmoidal

Packet Loss Linear, sigmoidal

Jitter Linear, sigmoidal

Response Time Linear

Service Completion Time Linear, polynomial, exponential

In [GAPRS05] author makes use of a methodology based on fuzzy-neural systems in order to carry out a coordinated management of the radio resources among the dierent access networks. In [Zha04], the author uses fuzzy logic to deal with imprecise criteria and user preferences; data are rst converted to numbers and then classical Multiple Attribute Decision Making (MADM) methods as Simple Additive Weighting (SAW) and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), are applied. Another approach aims to prioritize the available RATs to decide the optimum one for mobile users. Such approach was applied in [SJ05], using Grey Relational Analysis (GRA), which

(31)

2.3. Access Selection Decision Making in Heterogeneous Networks aims to prioritize the networks for the selection decision, after dening an ideal solution. Analytical Hierarchy Process (AHP) was adopted to arrange the decision parameters in three hierarchical levels, in order to calculate the corresponding weighting factors. Another exemple of combining GRA with AHP-based weighting is presented in [WK13]. Figure 2.2 describe how AHP can be used in order to calculate the decision parameters weights, then use them in order to calculate networks coecients in GRA and make the selection decision.

NPH approach, introduced in [KJ12b], consists of dening the SAW score for the ideal solution, calculates the SAW score for every candidate, and then computes the distances of each candidate score to the ideal solution score. Finally, the access network with the closest score to the ideal one is selected for the service. The ideal solution score is the user's SAW score considering the QoS parameters required by the user's application. In [CMT08], authors use AHP and GRA in order to construct the payo of requests and achieve network selection using multi-round game.

Figure 2.2: MADM with AHP-based weighting

In [SNW06], a performance comparison was made between Multiplicative Exponent Weighting (MEW), SAW, TOPSIS and GRA. Results showed similar performance to all trac classes. However, higher bandwidth and lower delay are provided by GRA for interactive and background trac classes. A network centric approach is adopted in

(32)

2.3. Access Selection Decision Making in Heterogeneous Networks [Tah07], to ensure load balancing, while minimizing the costs of resource underutilization and demand rejection.

In our work, we exploit the advantages of MADM techniques and especially the sim-plicity of SAW and NPH to develop a novel decision algorithm. Chapter 3, introduces our proposed algorithm, NP-BPA, for the access selection in a multi-operator network environment [FSL14, FSLC14]. Further, in the same chapter a comparison of our algo-rithm with SAW and NPH is made, results show the eciency of our decision algoalgo-rithm a three operators sharing network [FSLC15].

2.3.2 Access selection decision making in a Multi-Operator

network

In a multi-operator heterogeneous network, a new ex service paradigm was introduced

in [FPK+_{12]. It allows a mobile user subscribed to Flex service to dynamically}

ac-cess base stations (BSs) of dierent providers based on various criteria, such as prole, network conditions and oered prices. Flex users can select the appropriate provider and BS on a per-session basis. Authors present two modeling framework for the access markets at both microscopic and macroscopic levels. At a macroscopic level, users are considered as a homogeneous population with respect to preferences and decision-making mechanism. The behavior of users is described by a population game in order to deter-mine how the entire user population reacts to the decision of providers. At a microscopic level, a ex user accesses dynamically base stations of dierent providers based on various criteria, such as prole, network conditions and oered prices. At this level the model-ing framework and simulation platform are based on dierent modules concernmodel-ing the providers, the clients and the u-map, that serves as a review/feedback system from users and providers. The overview of the main modules of the microscopic level framwork is presented in Fig. 2.3. The client module contains information about the user service choice, the selected BS, its prole...The user prole determines the user constraints on cost, blocking probability and data rates, in addition to its preferences. And, the provider modules contains the price adaptation and the network blocking probability estimation. In our work, we envisage a similar multi-operator environment, where a user can access the base station of a dierent provider. However, our considered market is more open than Flex service market, since a mobile user does not need any previous subscription as a Flex user. Besides, the access selection decision is controlled by the home operator; the user is not free to switch between operators.

(33)

2.3. Access Selection Decision Making in Heterogeneous Networks

Figure 2.3: An overview of the main modules of the Flex access market

In fact, the majority of the existing works, in multi-operator environment, use game theory for the access selection and the joint service pricing. In [CSMW02], authors applied a non-cooperative game that makes use of Leaderfollower model (Stackelberg game) in order to study the competition between two ISPs. With a simple QoS model, a Nash equilibrium point was found from which the two ISPs would not move without cooperation.

In [KCG09], game theory is used for Dynamic Spectrum Access algorithm with cellular operators. Authors have dened a utility function, for the operators, considering user's bit rate, the blocking probability and the spectrum price. Moreover, they have presented a penalty function to control the blocking probability.

In cognitive radio networks [EMCA13], where mobile users may switch in real time to the provider (or providers) oering the best tradeos in terms of QoS and paid price, Nash equilibrium concept is used to nd the optimal price in a Stackelberg game between primary and secondary operators and Wardrop equilibrium is determined for the network selection game. Authors reveal the advantage for the primary operator to play before the secondary operator, particularly in a high-trac regime.

Furthermore, a two-stage multi-leader-follower game is used to model the interaction of a number of wireless providers and a group of atomic users in [GHR14]. The providers announce the wireless resource prices in a rst stage and the users announce their demand for the resource in the second stage. The user's choice is based on provider's prices and its channel conditions. Authors showed that the provider competition leads to a unique socially optimal resource allocation for a broad class of utility functions and a generic

(34)

2.4. Service Pricing in a Multi-Operator Network channel model.

In this thesis, chapter 4, we modeled the interaction between wireless operators, in a multi-operator sharing network, as a multi-leader-follower (Stackelberg) game. Cooper-ating service operators announce their transaction cost in the rst stage and the home operator of the transferred user performs the selection decision in the second stage. The game solution is found using Nash equilibrium concept, and the best response is determine

for every pairs of leaders [FCS+_15].

Another approach for Joint Radio Resource Management (JRRM) is introduced in [GAPRS07, GAPRS08]. Authors extended their single operator approach to a cooper-ation scenario between operators. They proposed a two-layer JRRM strategy to fully exploit the available radio resource and to improve operator revenue. The proposed economic-driven JRRM is based on fuzzy neural methodology with dierent classes of input parameters: technical inputs, economic inputs and operator policies.

Furthermore, a comparison between dierent access selection techniques was made in [WK13], it shows the strong and weak points of each techniques. We resume the comparison results in Table 2.3. We can point out on the implementation simplicity of MADM and its high precision, in addition to the high precision of game theory and its ability to fulll an equilibrium between multiple entities. This made game theory the rst choice to use in a sharing networks, where the partners seeks to selshly maximize their gains.

Table 2.3: Comparison of Network Selection Techniques

2.4 Service Pricing in a Multi-Operator Network

In multi-operator networks, the mainstream models suggested in litterature consider a pricing game between radio access network operators [NH08, ZZ13, ZHN14, ZBDH14,

BKA+_{15, BKA}+_{13, AKB}+_{15, FPK}+_{12]. Commonly, the mobile user is a player of the}

game, his strategy is to select the best access that maximizes his own utility. The latter is a function of the available QoS and the access price. It is assumed that the user has to

(35)

2.4. Service Pricing in a Multi-Operator Network pay for his new access, and the access price is decided dynamically by the operators or service providers, in order to maximize their payos. In this case, a competitive pricing scheme using hierarchical Stackelberg game is adopted.

In [ZBDH14], authors propose a multi-leader multi-follower Stackelberg game between Wi-Fi, small cell service providers (SCSP), macrocell service providers (MSP) and mobile users. Wi-Fi, SCSP and MSP are considered as the leaders of the game, and has to decide their service prices. The mobile users are the followers and based on the price of all the leaders, they selects a mixed strategy and chooses each leader with some probability. The utility of a leader represents its revenue, and the utility of the follower is the spectrum eciency of an access minus the price to be paid. In addition, when a user is served by SCSP, he has to pay a second price for the MSP as an interference penalty price. The best price to adopt is found using Stackelberg equilibrium notion, by solving a multi-objective two-levels optimization problem.

In a similar oligopoly market, studied in [ZZ13], a number of wireless Access Points (APs), controlled by dierent Service Providers (SPs), compete for the service of large number of end users. The SPs as leaders set prices for APs rst; and the end users as followers decide whether to accept the services and if they do, further decide which access point to select. In fact, users must pay before use. Authors dene a disutility of accessing an AP which is the sum of the price set by this AP and the congestion function. The user would access the APs with least disutility. In the adopted system model, the number of end users is assumed large and the impact of a single user on the whole system is negligible, thus authors used Wardrop principle to nd the equilibrium distribution of user demand ows on all APs. Besides, the payo of a leader is dened as the Prots of an AP/SP, and the oligopoly equilibrium for the access price is achieved at the Stackelberg equilibrium point, that maximizes the SP individual prots given the ow distribution of the end users. The SP prot-maximizing problem is solved analytically when there only exist two APs, and when there are more APs the problem turns to be complex and intractable. In this work, there is no explicit cooperation cost interchanged between SPs, and the end users are considered homogeneous for all APs.

Although, in [ZHN14], the cooperation between MSP and SCSPs is studied, the small cell networks (SCNs) are assumed to operate in a hybrid access mode. Authors address the radio resource sharing and the service price that a MSP pays to SCSP, in addition to the service selection performed by the users. The behavior of users is qualied as dynamic and time-varying with the performance satisfaction level and cost. Thus, the need of MSP and SCSP to dynamically adjust the price and the open access ratio (resource sharing ratio), in order to match the time-varying demand of the users. For this objective, authors used a hierarchical dynamic game framework based on dierentiel game theory and evolutionary game theory, in order to capture the dynamic behavior of SP and users. At the low level, the dynamic service selection is formulated as an evolutionary game, and the solution is found using Evolutionary Stable Strategy (ESS ), the user makes the

(36)

2.4. Service Pricing in a Multi-Operator Network service selection decision according to the access price and received throughput. In the upper level, the MSP and SCSPs sequentially determine the optimal pricing strategy and the open access ratio, using a Stackelberg dierential game. The MSP as leader oers a service price to SCSPs in order to aect their open access ratio. The SCSPs, the followers control the open access ratio to maximize their own payos. The solution of the game is found using open-loop Stackelberg equilibrium. In the proposed framework, the user pays a xed access price and receive a time-vatying throughput aected by the open access ratio, and the cooperation fees are paid by the MSP as a unit price for shared bandwidth. A similar rewarding framework is proposed in [SPTC15], author uses a two-stages sequential game between MSP and femtocell owners. A reward is oered by the MSP to the femtocell owners in order to aect the spectrum sharing ratio. This reward is a share of the revenue from hybrid users payment. These users are players at the second stage of the game, they perform service selection decision between Macro or Femto-service. The advantage of these framworks is that a user is out of the pricing game and a xed price is kept for his access. Such framwork can be adapted in a multi-operator environment, where we assume that the user always pays the access/service price set by his home operator. And when a transaction is performed, the home operator of the considered user pays a transaction cost to the new service operator. The latter sets its transaction cost and may control the portion of their shared resources, in order to maximize its prots.

Another approach excludes the users from the service pricing game. They are charac-terized by their demand and its probability distribution. The operators/service providers compete in order to maximize their prot in an oligopoly market. [NH08] represents two competitive pricing models for WiMAX and WiFi-based heterogeneous wireless network. The interaction between SPs was modeled using non-cooperative game models. Authors considered the case where SPs decide their price in a simultaneous-play, in this case, the solution is given by the Nash equilibrium. And, the case where WiMAX SPs are the leader and oer their prices before WiFi in a leader-follower game, in this case the solution in given by the Stackelberg equilibrium.

A two-stage multi-leader multi-follower game, called data ooading game is intro-duced in [GIHT13]. It models the interactions between BSs and APs in a free market as a two-stage non-cooperative game. In the rst stage, every BS proposes the price that it is willing to pay to each AP for ooading its trac. In the second stage, every AP indicates the trac volume its is willing to ooad for every BS. BSs are considered as the leaders of the game and APs are the followers. Authors showed that under the Nash equilibrium, every AP accepts only the BSs proposing the highest price, this price results from equalizing the maximum marginal cost reduction of all BSs and the marginal payment to the AP. In the proposed model, authors focused on the cost reduction of BSs and the prot improvement of the AP achieved from data ooading. The BS operator proposes the price as a reward for ooading the BS trac.

(37)

2.5. Conclusion serving a portion of their demand. This payment can be xed before the cooperation game or could be decided when interacting. Such model of base station sharing is presented in [LMK14], where authors built a simple microeconomic model that examines the behavior of base station operators who are collocated in a single cell. Authors used a game theoretic formulation, where BS operators interact to decide about turning on or o each BS in order to maximize the global utility. The latter is the summation of all operators utilities taking into account the demand distributions for each customer, their energy costs, revenue from a served customer, loss of revenue from dissatised customers, service capacity and payment rate.

In [BKA+_{15], a roaming-based infrastructure sharing scheme is proposed. The}

switch-ing o decision process is modeled usswitch-ing a static non-cooperative game played by N MNOs in M peripheral cells. Authors consider that part of the BS infrastructure in the M sur-rounding cells may be switched o during low trac conditions, motivating MNOs to share the resources of the remaining active BSs in the same cell. The switching o al-gorithm aims at minimizing the individual MNO cost in a distributed manner. Authors dened a cost function that explicitly considers the roaming and operational costs for MNOs. Such that, when the trac of a switched BS is roamed to an active BS, the MNOs of deactivated BSs pay a roaming cost to the active operators. The latters must consider additional cost for serving the roamed trac, and the roaming cost is considered as a portion of the total operational cost. The selection of the BS for trac roaming is made randomly with equal probability.

Our challenge is to design a framework for resource sharing and transaction cost pric-ing that involves access selection in a multi-operator environment, in order to guarantee operators and users satisfaction in the same time. None of the above sharing schemes proposes a solution on this issue.

2.5 Conclusion

In this chapter, we represented RAN sharing as a promising solution to upgrade mobile operators networks, in order to expand coverage, increase capacity, support higher data rates and enhance QoS in terms, with energy and cost eciency. In fact, Active RAN sharing will reduce the time and the cost of deploying new mobile technology. Besides, we made a review of the main RAT selection methods, and classied them into mono-operator access network and multi-mono-operator access network selection decision. In a shared RAN, access selection decision for multi-operator networks is needed, and the majority of works adopt game theoratical approaches to model the interaction between the dierent operators. Moreover, we outlined the principal approaches for inter-operators service pricing in a multi-operators/service providers environment. This service price can be a share of the user payment or a xed price. In addition, it may be determined by the cooperating service provider or may be a reward from the home operator of the user

(38)

2.5. Conclusion to this service provider. Including, we represented dierent models, where the user is a player and has to pay the cooperation fees, and other models where the user pays a xed price and its the home operator pay these fees. In the following chapter, we introduce a new hybrid decision method for the access selection in a multi-operator environment, that maximizes jointly the operators and users satisfaction.